LGTM

LGTM, but I need someone to look at the rust code here too

Nice! I'd love an experimental UI for this if this is the last py2 release, but even if it's just used in the background so that rebasing/amending added files stops dropping these bits, it would be a major benefit.

LGTM, but I still get an import error for this. Since there are a bunch of accepted things not pushed yet, maybe this has a dependency on one of those things.

This seems fine, but I couldn't import it without the previous commit, and I'd like someone who knows Rust to handle that.

In D11449#175515, @marmoute wrote:

If someone on centos can test this, it would be great.

In D11308#173761, @marmoute wrote:

In D11308#173760, @mharbison72 wrote:

In D11308#173544, @marmoute wrote:

This looks good (i.e. similar to what is done in the ./hg script).
Do you think we could get a test for that in test-extensions ? (using a fake APPDATA for that part of the test) ?

I haven't quite got the test working yet (should foo.py in site-packages be loadable, or does it need a directory with __init__.py?). But I see that site.getuserpackages() is available on py3, and it would be easier to write a test against that with support for all platforms. But that should (probably?) go on default, especially since the non pyoxidizer path needs to be updated to stay in sync. So how about this for stable, and a rework on default that supports py3 everywhere.

You mean getting this specific patch for stable, without a test, and a more complete version for default, with a test ?

In D11308#173544, @marmoute wrote:

This looks good (i.e. similar to what is done in the ./hg script).
Do you think we could get a test for that in test-extensions ? (using a fake APPDATA for that part of the test) ?

In D11307#173445, @marmoute wrote:

I seems a bit weird to be that we try to manually put self._bindir and self._hgcommand together while if I am not mistaken, they both originate from the splitting of some absolute path somewhere. So keeping that absolute path around would seems safer/cleaner to me. However I might be missing something.

Not sure that we want to take this as-is (even though it does avoid a bad problem outside of the internal tests). I'm hoping someone has some insight and saves me some time trial and erroring through this.